Detection of two-sided alternatives in a Brownian motion model (0704.3094v1)

Abstract: This work examines the problem of sequential detection of a change in the drift of a Brownian motion in the case of two-sided alternatives. Applications to real life situations in which two-sided changes can occur are discussed. Traditionally, 2-CUSUM stopping rules have been used for this problem due to their asymptotically optimal character as the mean time between false alarms tends to $\infty$. In particular, attention has focused on 2-CUSUM harmonic mean rules due to the simplicity in calculating their first moments. In this paper, we derive closed-form expressions for the first moment of a general 2-CUSUM stopping rule. We use these expressions to obtain explicit upper and lower bounds for it. Moreover, we derive an expression for the rate of change of this first moment as one of the threshold parameters changes. Based on these expressions we obtain explicit upper and lower bounds to this rate of change. Using these expressions we are able to find the best 2-CUSUM stopping rule with respect to the extended Lorden criterion. In fact, we demonstrate not only the existence but also the uniqueness of the best 2-CUSUM stopping both in the case of a symmetric change and in the case of a non-symmetric case. Furthermore, we discuss the existence of a modification of the 2-CUSUM stopping rule that has a strictly better performance than its classical 2-CUSUM counterpart for small values of the mean time between false alarms. We conclude with a discussion on the open problem of strict optimality in the case of two-sided alternatives.